diffusers

diffusers copied to clipboard

Dreambooth example on SD2-768 model is producing weird results

16

### Describe the bug Hi, This only happens on the 768 model, the base and 1.5/1.4 are not affected as far as I can tell. Here are the samples produced...

devilismyfriend

Fix dtype model loading

1

Makes sure weights are always loaded in the correct format

patrickvonplaten

[Dreambooth Example] Attempting to unscale FP16 gradients.

18

### Describe the bug I had the training script working fine but then I updated diffusers to 0.7.2 and now I get the following error: ```sh Traceback (most recent call...

jpiabrantes

For MPS using num_images_per_prompt with StableDiffusionImg2ImgPipeline results in noise

12

### Describe the bug If you try to generate multiple images with StableDiffusionImg2ImgPipeline by using the num_images_per_prompt parameter, under MPS all you get is noise/brown images. ![CleanShot 2022-10-22 at 10...

FahimF

how to combine train_unconditional.py with LDMPipline

1

### Describe the bug how to combine train_unconditional.py with LDMPipline ### Reproduction _No response_ ### Logs _No response_ ### System Info diffuser=0.6.0 python=3.8.12 pycharm

panenyaun

[Announcement] Stable Diffusion v1-4/v1-5 have lifted gated access! => No need to be logged into hf.co anymore!

2

https://huggingface.co/CompVis/stable-diffusion-v1-4 https://huggingface.co/CompVis/stable-diffusion-v1-3 https://huggingface.co/CompVis/stable-diffusion-v1-2 https://huggingface.co/CompVis/stable-diffusion-v1-1 are all no longer gated! This means you don't have to be logged in to download the model. Just doing: ```python from diffusers import DiffusionPipeline pipeline...

patrickvonplaten

Where to save the model locally?

1

I cannot believe that I really ask this but after two hours of research I didn't find a working way. I did it very often a few months ago but...

Marcophono2

[OOM] Memory blows out when trying to upscale images larger than 128x128 using StableDiffusionUpscalePipeline

12

### Describe the bug When trying to upscale images larger than 128x128 the progress goes to 100% and then crashes with CUDA OOM. With 512x512 images it's trying to allocate...

qunash

Fix Flax flip_sin_to_cos

2

Adding flip_sin_to_cos arg for FlaxTimesteps and passing it to get_sinusodial_embeddings. Currently, FlaxTimesteps hardcodes the order of the sin and cosine embeddings ([1](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/embeddings_flax.py#L92)). However, in the UNet2DConditionModel the order is an...

akashgokul

8k Stable Diffusion with tiled VAE

1

This PR makes it possible to generate 4k images in 8GB of VRAM using a tiled VAE codec combined with `enable_xformers_memory_efficient_attention()`. With 24GB of VRAM, you can generate 8k images....

kig